Skip to content

Qwen3 4B VRAM requirements

How much VRAM Qwen3 4B needs
4-bit weights: 2.3 GBWith 8K context: 4.2 GBWith 32K context: 7.6 GB8-bit, 8K context: 5.9 GB
Parameters
4B
Longest context
40K
Released
Apr 2025
License
Apache 2.0
Strongest at
Not rated

How much VRAM Qwen3 4B needs

Memory Qwen3 4B needs at each precision and context length
PrecisionWeightsWith 8KWith 32KWith 40K
16-bit7.5 GB9.4 GB12.8 GB13.9 GB
8-bit4.0 GB5.9 GB9.3 GB10.4 GB
6-bit3.1 GB5.0 GB8.4 GB9.5 GB
5-bit2.7 GB4.6 GB8.0 GB9.1 GB
4-bit2.3 GB4.2 GB7.6 GB8.7 GB
3-bit1.8 GB3.8 GB7.1 GB8.3 GB
2-bit1.4 GB3.3 GB6.7 GB7.8 GB
1.6-bit791 MB2.7 GB6.1 GB7.2 GB

One conversation, 16-bit context cache, runtime overhead included.

GPUs and Macs that run Qwen3 4B

The best way Qwen3 4B runs on common GPUs, Macs and AI boxes
MachineBest fitContextSpeedFirst replyReport
RTX 40608-bit16K43 tok/s0.5 sPDF: Qwen3 4B on RTX 4060 report
RTX 50608-bit16K68 tok/s0.4 sPDF: Qwen3 4B on RTX 5060 report
RTX 30708-bit16K68 tok/s0.4 sPDF: Qwen3 4B on RTX 3070 report
RTX 3080 10 GB8-bit32K110 tok/s0.3 sPDF: Qwen3 4B on RTX 3080 10 GB report
RTX 3060 12 GB8-bit40K55 tok/s0.6 sPDF: Qwen3 4B on RTX 3060 12 GB report
Arc B5808-bit40K48 tok/s0.4 sPDF: Qwen3 4B on Arc B580 report
RTX 40708-bit40K76 tok/s0.3 sPDF: Qwen3 4B on RTX 4070 report
RTX 4070 Super8-bit40K76 tok/s0.2 sPDF: Qwen3 4B on RTX 4070 Super report
RTX 50708-bit40K98 tok/s0.3 sPDF: Qwen3 4B on RTX 5070 report
RTX 4060 Ti 16 GB8-bit40K45 tok/s0.4 sPDF: Qwen3 4B on RTX 4060 Ti 16 GB report
RTX 5060 Ti 16 GB8-bit40K68 tok/s0.3 sPDF: Qwen3 4B on RTX 5060 Ti 16 GB report
Arc A770 16 GB8-bit40K57 tok/s0.4 sPDF: Qwen3 4B on Arc A770 16 GB report
Radeon RX 9070 XT8-bit40K77 tok/s0.2 sPDF: Qwen3 4B on Radeon RX 9070 XT report
RTX 4070 Ti Super8-bit40K98 tok/s0.2 sPDF: Qwen3 4B on RTX 4070 Ti Super report
RTX 4080 Super8-bit40K107 tok/s0.2 sPDF: Qwen3 4B on RTX 4080 Super report
RTX 5070 Ti8-bit40K127 tok/s0.2 sPDF: Qwen3 4B on RTX 5070 Ti report
RTX 50808-bit40K134 tok/s0.1 sPDF: Qwen3 4B on RTX 5080 report
Radeon RX 7900 XTX8-bit40K108 tok/s0.4 sPDF: Qwen3 4B on Radeon RX 7900 XTX report
RTX 30908-bit40K132 tok/s0.2 sPDF: Qwen3 4B on RTX 3090 report
RTX 40908-bit40K140 tok/s0.1 sPDF: Qwen3 4B on RTX 4090 report
Radeon AI PRO R97008-bit40K77 tok/s0.2 sPDF: Qwen3 4B on Radeon AI PRO R9700 report
RTX 50908-bit40K223 tok/s0.1 sPDF: Qwen3 4B on RTX 5090 report
2× RTX 30908-bit40K132 tok/s0.1 sPDF: Qwen3 4B on 2× RTX 3090 report
2× RTX 40908-bit40K140 tok/s0.1 sPDF: Qwen3 4B on 2× RTX 4090 report
RTX PRO 5000 Blackwell 48 GB8-bit40K178 tok/s0.1 sPDF: Qwen3 4B on RTX PRO 5000 Blackwell 48 GB report
RTX 6000 Ada8-bit40K134 tok/s0.1 sPDF: Qwen3 4B on RTX 6000 Ada report
L40S8-bit40K123 tok/s0.1 sPDF: Qwen3 4B on L40S report
2× RTX 50908-bit40K223 tok/s0.1 sPDF: Qwen3 4B on 2× RTX 5090 report
4× RTX 30908-bit40K132 tok/s0.1 sPDF: Qwen3 4B on 4× RTX 3090 report
RTX PRO 6000 Blackwell8-bit40K223 tok/s0.1 sPDF: Qwen3 4B on RTX PRO 6000 Blackwell report
Mac mini M4 16 GB8-bit36K22 tok/s5.2 sPDF: Qwen3 4B on Mac mini M4 16 GB report
Mac mini M4 24 GB8-bit40K22 tok/s5.2 sPDF: Qwen3 4B on Mac mini M4 24 GB report
Mac mini M4 32 GB8-bit40K22 tok/s5.2 sPDF: Qwen3 4B on Mac mini M4 32 GB report
Mac mini M4 Pro 24 GB8-bit40K46 tok/s2.6 sPDF: Qwen3 4B on Mac mini M4 Pro 24 GB report
Mac mini M4 Pro 48 GB8-bit40K46 tok/s2.6 sPDF: Qwen3 4B on Mac mini M4 Pro 48 GB report
Mac mini M4 Pro 64 GB8-bit40K46 tok/s2.6 sPDF: Qwen3 4B on Mac mini M4 Pro 64 GB report
M4 Max 128 GB8-bit40K75 tok/s1.3 sPDF: Qwen3 4B on M4 Max 128 GB report
M5 Max 128 GB8-bit40K82 tok/s0.4 sPDF: Qwen3 4B on M5 Max 128 GB report
M3 Ultra 96 GB8-bit40K92 tok/s0.7 sPDF: Qwen3 4B on M3 Ultra 96 GB report
M3 Ultra 256 GB8-bit40K92 tok/s0.7 sPDF: Qwen3 4B on M3 Ultra 256 GB report
M3 Ultra 512 GB8-bit40K92 tok/s0.7 sPDF: Qwen3 4B on M3 Ultra 512 GB report
M5 Ultra 512 GB8-bit40K120 tok/s0.2 sPDF: Qwen3 4B on M5 Ultra 512 GB report
Ryzen AI Max+ 395 64 GB8-bit40K34 tok/s0.8 sPDF: Qwen3 4B on Ryzen AI Max+ 395 64 GB report
Ryzen AI Max+ 395 128 GB8-bit40K34 tok/s0.8 sPDF: Qwen3 4B on Ryzen AI Max+ 395 128 GB report
RTX Spark8-bit40K47 tok/s0.3 sPDF: Qwen3 4B on RTX Spark report
DGX Spark8-bit40K43 tok/s0.3 sPDF: Qwen3 4B on DGX Spark report
Dual DGX Spark8-bit40K43 tok/s0.2 sPDF: Qwen3 4B on Dual DGX Spark report
DGX Station GB3008-bit40K515 tok/s0.1 sPDF: Qwen3 4B on DGX Station GB300 report
2× RTX PRO 6000 Blackwell8-bit40K223 tok/s0.1 sPDF: Qwen3 4B on 2× RTX PRO 6000 Blackwell report
4× RTX PRO 6000 Blackwell8-bit40K223 tok/s0.1 sPDF: Qwen3 4B on 4× RTX PRO 6000 Blackwell report
8× RTX PRO 6000 Blackwell8-bit40K223 tok/s0.1 sPDF: Qwen3 4B on 8× RTX PRO 6000 Blackwell report
A100 80 GB SXM8-bit40K245 tok/s0.2 sPDF: Qwen3 4B on A100 80 GB SXM report
H100 SXM8-bit40K345 tok/s0.1 sPDF: Qwen3 4B on H100 SXM report
4× H200 NVL8-bit40K425 tok/s0.1 sPDF: Qwen3 4B on 4× H200 NVL report
8× H200 NVL8-bit40K425 tok/s0.1 sPDF: Qwen3 4B on 8× H200 NVL report

Best fit is the highest quality, up to 8-bit, that leaves 16K of context. First reply is the wait for a 2K prompt. Speeds are rough, plus or minus 25%.

How to run it

llama.cpp
llama-server -hf Qwen/Qwen3-4B-GGUF
vLLM
vllm serve Qwen/Qwen3-4B
MLX (Mac)
mlx_lm.chat --model mlx-community/Qwen3-4B-4bit

Questions

How much VRAM does Qwen3 4B need?

Qwen3 4B needs 2.3 GB for its weights at 4-bit, 4.2 GB with an 8K context and 7.6 GB with 32K. At 8-bit, 5.9 GB.

What is the minimum VRAM for Qwen3 4B?

About 3.8 GB at 3-bit with an 8K context, the lowest precision we recommend for a model this size. The smallest machine here that runs it without spilling is the RTX 4060.

How big is Qwen3 4B?

4B parameters: 7.5 GB of weights at 16-bit, 2.3 GB at 4-bit.

Can Qwen3 4B run on an RTX 4090?

Yes. At 8-bit it holds 40K of context and writes about 140 tokens a second.

More models